Tracking of Monthly Health Condition Change from Daily

Measurement of Systolic Blood Pressure

Wenxi Chen

and Toshiyo Tamura

Biomedical Information Technology Lab., CAIST, The University of Aizu, Tsuruga, Aizu-wakamatsu 965-8580, Japan

Faculty of Biomedical Engineering, Osaka Electro-Communication University, Shijonawate, Osaka 575-0063, Japan

Keywords: Health Condition, Biorhythm, Long-term Monitoring, Monthly Change, Daily Measurement, Systolic Blood

Pressure, Healthcare, Dynamic Time Warping.

Abstract: This paper presents an approach to detect monthly biorhythmic change using daily measurement of systolic

blood pressure (SBP) at home. As a part of health promotion campaign initiated in 1994, more than 600

households in West Aizu village of northern Japan were provided devices for daily measurement of blood

pressure, electrocardiogram, body temperature and body weight. This paper demonstrates an outcome of

data analysis of daily SBP collected in two years from an elder couple at age of seventies. The personal

reference profile is gained by averaging individual monthly profiles over 24 months. Dynamic time warping

algorithm estimates the similarity between personal reference profile and monthly SBP profile. The results

show that an extraordinary deviation from usual biorhythmicity can be found in both the wife and the

husband happened in July and February which respectively indicates individual health condition change

confirmed by personal medical record. The results suggest that even it is difficult to identify any significant

variation from the daily SBP directly, proper analysis of the raw SBP measured over a long-term period

helps tracking functional information of health condition change and serving as an effective evidence for

health management.

1 INTRODUCTION

Flood of information brings a big impact on the way

we live and work. Every day, 2.5 quintillion bytes of

data are being generated, and so much that 90% of

the data in the world today have been created in the

last two years alone (IBM Corp., 2011). These data

come from everywhere such as sensors, posts,

pictures and videos, transaction records and personal

information. Increase in quantity philosophically

will lead to profound change in quality. The vast

amount of data is more than simply a matter of size,

and sometimes is likely a double-edged sword. It

usually has a huge reserve of latent information but

often blurs the focus of the interests.

It is crucially an important challenge in exploring

proper approaches to handle these data and to mine

functional information from daily accumulated such

kind of data, and ultimately to discover structural

knowledge for real world application (Zins, 2007).

Detection of influenza epidemics using only

search engine query data announced the arrival of

the Big Data age and paved the way for finding new

value from multiple disciplines (Ginsberg et al.,

2008).

Diversified devices were developed to acquire

multifarious physiological data under daily life

environment conveniently. Variety of algorithms

were devised to reveal the relationship between data

features and physiological signatures in healthcare

domain.

West Aizu village, located in northern Japan and

about 300 km away from Tokyo, had pioneered the

“Challenge to 100 years of age” project since 1994.

The project had been supported by various financial

resources of total 2.4 billion Japanese Yen, and

established its fundamental goal to promote healthier

life by providing a total care solution package to

villagers (West Aizu, 2003). The village built a

cable television network infrastructure, improved the

soil for the cultivation of crops, enhanced

educational programs on the importance of a

nutritionally balanced diet and good lifestyle

practice, and initiated a health promotion campaign.

Special tailor-made devices were distributed to 687

households among total 2,819 families in the village.

Daily physiological data are measured by

Chen W. and Tamura T..

Tracking of Monthly Health Condition Change from Daily Measurement of Systolic Blood Pressure.

DOI: 10.5220/0005203400690074

In Proceedings of the International Conference on Health Informatics (HEALTHINF-2015), pages 69-74

ISBN: 978-989-758-068-0

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

participants at home and transmitted from home to

the healthcare centre via the cable network.

This paper is to explore a feasible way to harvest

such kind of daily data accumulated over a long-

term period, and to find structural and functional

information which can be linked to health condition

change.

2 METHOD AND MATERIAL

2.1 Data Collection

The tailor-made device can measure systolic blood

pressure (SBP), diastolic blood pressure (DBP),

body temperature (BT), body weight (BW), one-

minute electrocardiogram (ECG) and heart rate (HR)

profile, and also collect answers to a daily

questionnaire displaying on a LCD screen after

completion of daily measurement. Measured data are

transmitted to the healthcare centre by home

network connection and accumulated in the database

server of the centre. The time of the daily

measurement is not strictly stipulated: preference of

the morning or the afternoon is at the disposal of the

participants. Seven nurses are in charge of the data

review and respond to inquiries from the

participants. Biochemical markers from blood and

urine samples are also collected in yearly regular

health check-up.

The participants involved in the project were

given the explanation on the study purpose and the

daily tasks, and were asked to sign an agreement

prior to the data collection.

Figure 1: Snapshot in measurement of blood pressure

using the device at home. The measured data were

transmitted to a database server via a village network

connection.

Figure 1 shows a housewife measuring the blood

pressure at home by the device. Three large buttons

“Yes”, “No”, “Return” and a speaker for voice

guidance are designed especially for elders to

manipulate more easily.

Figure 2 shows daily measurements of SBP,

DBP, HR, BT and BW in two years from an elder

couple; upper and lower plots indicate the wife and

husband, respectively. The wife was born in 1925

and suffered from hypertension and had accepted

coronary artery bypass grafting surgery. The

husband was born in 1924 and had no overt

symptoms.

It is observable that the physiological data in the

wife demonstrates the wax and wane corresponding

to the temporal ebb and flow. SBP, DBP and HR

tend to decline in the summer and rise in the winter.

However, the biorhythmicity in the husband shows

an obscure pattern. The polynomial fitted curves

(brown lines) for the SBP show that individual

Figure 2: Profiles of daily HR, SBP, DBP, BT and BW in

two years collected from a couple ([upper] wife and

[lower] husband). Sporadic blanks indicate no

measurement on those days. The brown lines are the

polynomial approximation of SBP (9th order for the wife

and 3rd order for the husband). Most data were measured

in the afternoon of a day.

HEALTHINF2015-InternationalConferenceonHealthInformatics

biorhythmicity differs in period and MESOR,

amplitude and phase, zenith and nadir. Therefore,

proper analysis and visualization of these data are

indispensable in connecting daily physiological data

to biorhythmicity and health condition.

2.2 Pre-processing of SBP

This paper uses only the daily SBP data from the

above measurement in data analysis. Raw SBP data

are often contaminated by spike-like noise and other

artefacts due to poor contact and motion. The former

is suppressed by a median filter and the latter is

mitigated by a Savitzky–Golay filter.

2.2.1 Suppression of Spike-like Noise

Occasional spike-like SBP data are considered

outliers and suppressed in the first step.

A median filter is a nonlinear digital filtering

technique, usually used in the image processing field

to remove speckle noise, or salt/pepper noise from

images. The idea is to represent the signal by

replacing an extremely large value with a reasonable

candidate value. This is realized using a window

consisting of an odd number of data. The values

within the window are sorted in numerical order, and

the median value, the sample in the centre of the

window, is selected as the output of the filter.

When the window is moved along the signal, the

output of the median filter y(i) at a moment i is

calculated as the median value of the input values

x(i) corresponding to the moments adjacent to i

ranging from –L/2 to L/2.



































2/,12/

,...,,...,12/,2/

LixLix

ixLixLix

medianiy

(1)

where L is the length of the window.

2.2.2 Smoothing of Monthly SBP Profile

The Savitzky–Golay filter is used to smooth the data

outputted from the median filter. The Savitzky–

Golay filter segments the data as frames using a

moving window, and approximates the data frames

one by one using a high-order polynomial, typically

quadratic or quartic (Savitzky & Golay, 1964).

For each input point y(i), a digital filter output

z(i) can be expressed by a linear combination of the

nearby input points as

 









kiyciz

(2)

where n

is the number of points on the left-hand

side of the data point i, and n

is the number of

points on the right-hand side of i.

The Savitzky–Golay filter is to find a proper

polynomial to fit all n

+1 points within each

window frame on the least-squares meaning, and to

produce a filter output z(i) as the value of that

polynomial at position i.

To derive filter coefficients, c

, we consider

fitting a polynomial of degree M in i, namely

i+a

+•••+a

to the values y

−nL

,...,y

. Then,

z(0) will be the value of that polynomial at i = 0,

namely a

. The design matrix for this problem is

iA 

Mjnni

,...,0,,...,0,..., 

(3)

The normal equations for the polynomial

coefficients vector, a=[a

, a

,•••, a

]’, in terms of

the input data vector, y=[y

−nL

,...,y

]’, can be written

in a matrix notation as below:

yaA





(4)

The polynomial coefficients vector, a, becomes







yAAAa 



(5)

We also have the specific forms













kjki

kAAAA

(6)

and









kkj

ykyAyA

(7)

Since the filter coefficient, c

, is the component

when y is replaced by the unit vector e

, we have



 



















eAAA

(8)

where –n

≤ k < n

When the filter coefficient vector c=[c

-nL

,…,c

]

is obtained using Equation (8), the data can be

smoothed using Equation (2).

2.3 Detection of Biorhythmic Change

Biorhythmic change is detected by the dynamic time

warping (DTW) algorithm (Salvador and Chan,

2007). DTW is an algorithm used to measure the

TrackingofMonthlyHealthConditionChangefromDailyMeasurementofSystolicBloodPressure

similarity between two data sequences that may

generally vary in temporal span and rhythmic tempo.

2.3.1 DTW Algorithm

The aim of DTW is to find the optimal alignment

between two given data sequences under given

criteria. The length of two sequences may differs

and varies. The reference sequence R={r

,...,r

}

with length M, and the test sequence T={t

,...,t

}

with length N, are shown in Figure 3. The value of

each black dot d

indicates the difference (distance)

between the reference sequence r

and the test

sequence t

, as described by Equation (9).

 

iiij

trjid 

, i=1, 2,…, M;

j=1, 2,…, N,

(9)

Figure 3: Dynamic time warping algorithm showing the

optimal path (red line) between the reference sequence and

the test sequence.

Thus, a two-dimensional NM distance matrix,

N×M

, is constructed where the element d

is the

distance between the i

data in the reference

sequence and the j

data in the test sequence.

As a similarity measure, the shortest path from

the start (the lower left-hand corner of the distance

matrix) to the end (the upper right-hand corner of the

distance matrix) of the data sequence must exist

among multiple possible paths.

The shortest path is determined using the

forward dynamic programming approach with a

monotonicity constraint.





kijk

PdP

min







(10)

where P

denotes the distance from the i

and the

data node to the terminating node.

The overall minimum distance, D(T, R), used as

the similarity measure for two sequences (a smaller

distance value indicates a higher similarity) is

determined from





, PRTD



(11)

2.3.2 Personal Reference Profile

A personal reference profile is created by averaging

individual 24 monthly SBP profiles and used as the

reference sequence in DTW calculation. Because the

number of days in a month differs from month to

month, and data loss unavoidably happens in daily

measurement, the length of the reference profile is

normalized to 30 days by resampling the daily

measured raw data.

Two personal reference profiles for the wife and

the husband are shown in Figure 4.

Figure 4: Personal reference profiles derived by averaging

24 monthly profiles of individual SBP data.

2.3.3 Biorhythmic Change Index

The personal reference profile is used as a reference

sequence to calculate the overall distance D from

monthly SBP profile using the DTW algorithm as

shown in Equation (11). The D value is considered

as a similarity measure describing the discrepancy

between personal reference profile and monthly SBP

profile, and serves as a biorhythmic change index

(BCI) reflecting the monthly biorhythmic change.

The smaller the value of BCI is, the more regular in

biorhythmicity and the less change in health

condition.

3 RESULTS

The monthly change of BCI, or the overall distance

between personal reference profile and individual

monthly SBP profile in two years is shown in Figure

5. The upper plot presents the outcome of the wife,

the lower plot is for the husband. The BCI value was

calculated by the DTW algorithm using the personal

Reference sequence

Test sequence

100

110

120

130

140

150

1 6 11 16 21 26

SBP (mmHg)

Day in a month

Husband Wife

HEALTHINF2015-InternationalConferenceonHealthInformatics

reference profile as a reference sequence and the

monthly SBP profile as a test sequence. The smaller

the value of the BCI is, the higher the similarity has

between the personal reference and monthly profile.

Figure 5: Monthly change of the BIC, or the overall

distance between personal reference profile and monthly

SBP profile in the wife (upper plot) and the husband

(lower plot) in two years.

It is obvious that the monthly rhythmic change

can be found in both the wife and the husband.

Nevertheless, the exceptional change in the wife

happened in July, 2001 (blue trace in upper plot),

and husband in February, 2002 (red trace in lower

plot). The onset of the symptoms usually insinuates

certain alterations in physical or mental conditions.

This implies that remarkable change in the health

condition of the wife and the husband probably

occurred in the respective timing point.

The above analytical outcome were confirmed to

be accordant with the couple’s real situation by

referring to their personal medical records, both

were in poor health condition and were accepting

treatment in the corresponding period.

In comparison with the noteworthy biorhythmic

variations in July, 2001 of the wife and in February,

2002 of the husband, indistinctive change in other

corresponding months in two years exhibits a

monthly repetitive pattern of biorhythmicity in good

health condition.

4 DISCUSSION

Many chronic diseases, such as diabetes,

hypertension, arteriosclerosis, malignant neoplasm,

cerebral and cardiovascular conditions, are silent

killers that require a long-term course in disease

development and threaten human beings in a latent

way. Occasional or regular yearly health check-up

are difficult to identify the onset of the symptoms

and incidence of diseases at their early stage.

Various home-based devices provide convenient

approaches for daily measurement of variety of

physiological data in daily life environment.

Nevertheless, a huge volume of data accumulated

over a long-term period usually contain abundant

functional information but require proper approach

in order to assess the significant signature in

different physiological and pathological conditions.

As is well known, physical and mental

conditions are affected by various endogenous and

exogenous factors. They may include emotional,

psychological, behavioural aspects, and as well the

meteorological, environmental, geographical, and

temporal factors. Therefore, it is difficult to identify

the underlying regulatory mechanism which is

responsible for various benign and malignant

stimulants.

Instead of scrutinizing every detail of daily

measurement of SBP, we applied an efficient DTW

algorithm for tracking of biorhythmic alteration to

reflect the health condition change. The results also

suggest that it is possible to track not only

physiological condition change in monthly base but

also various specific events in daily base such as

heavy intake of alcohol, mental depression, and

other unusual incidents in daily life, provided the

vast amount of physiological data is accumulated

through daily measurement over a long-term period,

and proper algorithm is applied to scoop out the

valuable information.

Personal reference profile is currently obtained

by simply averaging all of the monthly data. It is

TrackingofMonthlyHealthConditionChangefromDailyMeasurementofSystolicBloodPressure

apparent that the aging process affects the averaged

personal reference profile, and evolution of the

personal reference profile is desirable to adapt on

monthly base gradually to reflect intrinsic

biorhythmic change with aging process in the future

study.

Although this paper presents only the outcome

obtained from two elders in two years, it is

promising to recognize the feasibility for tracking of

monthly change in health condition by daily

measurement of physiological data. More data from

more persons in different age groups, longer period

of measurement, and diversity of physiological and

pathological conditions are preferable in further

validation of the proposed method. More sensitive

and more robust algorithms are also worth to be

explored in depth on different temporal bases such

as daily, weekly, monthly, seasonal and yearly.

5 CONCLUSIONS

In this paper, we applied the DTW algorithm to

analyse the monthly rhythmicity using daily

measurement of SBP from an elder couple in two

years. Minor variation in monthly biorhythmicity

indicates the physiological adaptation to internal and

external factors temporally. The remarkable

deviation out of usual physiological adaptation

reflects the health condition alteration accordingly. It

suggests that the recognisable unusual change

provides an evidence to help making decision in

daily health management and an insight into chronic

disease control, and perception to deal with daily

health problem more smartly.

The results also suggest that even it is difficult to

identify any significant variation from the daily or

the monthly SBP profile directly, proper analysis of

the raw SBP measured over a long-term period is

able to help tracking functional information of health

condition change and serving as an effective

evidence for health management.

ACKNOWLEDGEMENTS

The authors would like to thank the volunteers for

their endurance in daily measurement of

physiological data over a long-term period. This

study is supported in part by the collaborative Grant

of Samsung Institute of Japan, No. A-26-1 and the

University of Aizu Competitive Research Funding,

No. P-27.

REFERENCES

IBM Corp., 2011. IBM big data and information

management http://www-01.ibm.com/software/data/

bigdata/ Accessed Aug. 22, 2014.

Zins, C., 2007. Conceptual Approaches for Defining Data,

Information, and Knowledge. Journal of the American

Society for Information Science and Technology;

58(4):479–493.

Ginsberg, J., Mohebbi, M., Patel, R., Brammer, L.,

Smolinski, M., Brilliant, L., 2008. Detecting influenza

epidemics using search engine query data. Nature;

457:1012-1014.

West Aizu, 2003. Challenge to 100 Years of Age – The

Making of a Healthy Village through a Total Care

Solution, Fukushima: Zaikai21 Publishing House.

Savitzky, A. & Golay, M. J. E. 1964. Smoothing and

Differentiation of Data by Simplified Least Squares

Procedures. Analytical Chemistry; 36(8):1627-1639.

Salvador, S., Chan, P., 2007. Toward Accurate Dynamic

Time Warping in Linear Time and Space. Intelligent

Data Analysis; 11(5):561–580.

HEALTHINF2015-InternationalConferenceonHealthInformatics